#GoogleIO has officially begun, and we hid our favorite offline companion in plain sight. Before we dive into the latest web platform updates, try and spot the Chrome Dino hidden in the crowd 👀

How does a game as pretty as Forza 6 have just the absolute worst character models of any game of this console generation?! Especially given how much screen time they get!

Learn how to measure human expert agreement with Cohen's Kappa, balance your judge's precision and recall using the F1 score, and avoid the massive trap of overfitting with a secret final exam dataset in our final AI evals segment → goo.gle/42BjEYU

Using an LLM to work through some systems design stuff and it's really giving Cunningham's Law. I mean I guess that means it's working?

This week marks the final sprint towards Google I/O next week … I’m very much looking forward to my talk “What’s new in Web UI” together with @una.im. io.google/2026/explore... Lots of tweaking and rehearsing, but we’re getting there :) (Yes, the talk is getting recorded) #GoogleIO #ChromeDevRel

Check out my new font, ANALOG MONO! Recreate 90's camcorder graphics or analog video on-screen displays! ► gleeson.itch.io/analog-mono

It's time to wire everything together into a continuous testing pipeline → goo.gle/4tSIcZo Today, we'll cover extended unit tests for PR merges, weighted Composite Metrics for nightly integration runs, and Pairwise Evaluation to mathematically prove which model is best when you want to swap LLMs.

We are still accepting developers into the WebMCP Early Preview program → goo.gle/41QD9MQ Teach AI agents how to navigate your site using the Declarative API for forms or the Imperative API for direct JavaScript access.

If anyone is interested in how great the Threads algorithm is. Viewer: he did not see fewer posts like this.

Building reliable AI tests require more than asking an LLM for a 1-10 score. By combining objective checks with an LLM judge that outputs a rationale and then a binary score, you get a much more robust pipeline. Learn the steps to set this up in part 4 and 5 of our AI evals series → goo.gle/4uDaaZl

Before you write any code, you need to define exactly what success and failure looks like for your AI application. Today, we hit the drafting board to map out objective failures and subjective ones → goo.gle/4tOVshb

i’ve worked in tech more than 20 years. it has never been more ridiculous.

We are bringing the latest web announcements to #GoogleIO on May 19th at 1:30pm PT to help you build incredible web applications → goo.gle/3OHt9Cm

We are back with the second installment of our AI testing guide → goo.gle/4tOVshb Today, @maudnals.bsky.social guides you through transitioning your standard web testing approach to handle the complexities of LLMs and subjective outputs 🧪

⭐ available for client work ⭐ Front-end developer & Design Systems consultant with 15 years experience building component libraries, modernising legacy codebases, writing docs & launching new projects Let's fix your tech debt or create something new [email protected] 🙏🖤

I really don't recommend chasing very spicy food with a frozen slush drink and getting brain freeze while your mouth is on fire.

Learn the science of AI evaluation with our new online series featuring @maudnals.bsky.social → goo.gle/4tOVshb We're sharing a scientific approach to building eval pipelines that ensure your app actually works for users 🧪

Roasted schmaltz vinaigrette. I will not be taking further questions. #cooking

Because at home printing is literally just a scam, I managed to buy a full home office printer/scanner combo yesterday for literally $30 cheaper than buying the ink, and almost half the price of a printer that had been discontinued and they were trying to drop inventory of.

Got a transatlantic flight today so I'm going to play around with running Gemma4' 26b model locally through Claude Code. If it works, this could be very fun and interesting... I'll keep you posted. #AI #OpenWeights

If you're into #ttrpg, I wanted a little tool to visualize dice probabilities based on ability score modifiers. Currently it's hard-coded to the percentages from Powered by the Apocalypse, but if someone's really interested I'll think of a way to add custom ranges. snugug.github.io/roll-probabi...

If you're into #ttrpg, I wanted a little tool to visualize dice probabilities based on ability score modifiers. Currently it's hard-coded to the percentages from Powered by the Apocalypse, but if someone's really interested I'll think of a way to add custom ranges. snugug.github.io/roll-probabi...

It has been so long since I have seen @snugug.bsky.social in person! Pictured here after oolong with @lawyerfox.bsky.social and me 💓

Tried the new vertical tabs in Chrome and I'm not sure they're for me y'all. It's nice having less window chrome, but I don't like the lost horizontal space when the tabs are open and I get lost when I close the tab sidebar. Why are y'all so gaga over it?

📣 After nearly 5 fulfilling years leading Core infrastructure organizations for Google, I’m thrilled to share that I’ve taken on a new role! I’m going back to IC work as a Distinguished Engineer- my new role is in Chrome as Area Tech Lead, AI Web Ecosystem! 💥

7 years at Google today!? Wow time flies. So much has changed but I'm still grateful to be doing meaningful work on a product that impacts billions of users with an awesome team 💙 This much hasn't changed 😊

Sharing again now that it's not a holiday! This weekend I pushed the first release of Nebula, a nearly 0-config Git-based CMS for @astro.build that builds itself from the Content Collections you've already written! github.com/snugug/nebul...

Gotta love how smart and totally actually intelligent the current spate of AI is: “ ~18GB at Q4 — tight for 16GB VRAM…”.

Hey @svelte.dev & @astro.build peeps: what's the best way to distribute Svelte components that actually keep their styling on build that's not injecting CSS via JS? I'm now on approach 3 which I think is "no CSS in Svelte, global files only". Current approach works in dev but not _entirely_ in prod

Hey everyone! I've got a new project to share with y'all! Introducing Nebula! A Git-based Content Management System designed for @astro.build ! github.com/Snugug/nebul... Nebula uses your existing Content Collection definitions to dynamically build the editing interface! No more double config!

Some flatpack furniture designers never played with LEGO as a kid _and it shows_.

Does anyone have a good IndexedDB backup/restore suggestion? #webdev #javascript

We could use your help. If you’ve ever wanted to work with us, now would be an excellent time! Please spread the word. Thank you! cloudfour.com/thinks/more-...

Why does using #ai feel magical for topics you don't know, but frustrating for those you do? Expertise still matter in the world of chatbots because true problem solving requires drawing on intangibles. Understanding LLM limitations is essential to getting good results. snugug.com/musings/ai-c...

Spec proposal merged. HTML video and audio lazy loading is now a web standard. github.com/whatwg/html/... html.spec.whatwg.org#the-video-el... html.spec.whatwg.org#the-audio-el...

Forgive me, gonna do some shameless reposting because a) aww.... and b) you might not realize that I'm open to fun, meaningful, *scoped and short-term* projects where I can help you help your docs/community! Not "hiring"? No problem! Maybe you just need someone to stay until the wind changes...

While I enjoy working with AI coding tools, it's very frustrating when they're confidently wrong—telling me an API can't do something when I did DevRel for it and I know it can. This is a key reason why I don't think AI coding tools are well suited for early-career devs.

I've been having a lot of conversations lately around #ai and #softwareengineering and #webdev and all of the conversations seemed to converge to the same thought: coding was never the job of a software engineer, it's the tool used to learn and practice systems design. snugug.com/musings/codi...

I've got a bone to pick with Hyundai. I recently got a plug in hybrid from them and I mostly drive on electric, but it's average MPG (L/100Km) keeps going DOWN because it's including in the average the 0 I use on electric! Technically correct but absolutely useless!

Take a bigger slice of the agentic web this #PiDay by shipping a literal pizza pie → goo.gle/3PaAVEL#PiDay To try: enable chrome://flags/#enable-webmcp-testing & install the Model Context Tool Inspector extension. Share what "pi" you’d build below! 🍕

I think my favorite bit of Astro 6 is the upgrade to Zod 4. This'll enable my absolute madness of using Astro's content types as the direct source of truth for a full CMS.

Rounds 1 and 2 of my desk decorations at work:

I apparently logged into my new Windows 11 install wrong and now the only way to have something resembling a normal computer experience is by reinstalling my OS from scratch. Windows Hello is the _worst_. Valve, make quick with the full Proton support for Windows games plz.

Partially related: I'm currently sitting in an airport lounge having Claude churn away at a piece of front end engineering I would have never tackled before because of the number of dependencies I'd have needed to learn and get right for a fun side project that I'd have never used again. 1/2

Just submitted my first conference talk in years: The Future of Software Engineering. A talk about what #AI coding is and isn't good at, how to get the most out of it, and how I think it'll change the software engineering practice. If you'd like this at your conference, reach out! #coding #webdev

Another thing Fred Brooks pointed out, half a century ago, is that a software design mistake that is easy and cheap to fix early, becomes more and more and more awful as time goes on. Catching shitty designs early can save *shit tons* of time and effort.

Meet Gemini 3.1 Flash-Lite: intelligent, high-volume AI with minimal latency → goo.gle/3OKbxFO Get the precision your dev workflows need without the overhead. Take 10 mins to try it in AI Studio and share what you built below!

Canada, question, where the heck did "antojitos" come from? I've lived in Texas, these aren't a thing, but they're THE "Mexican"/ "Tex-Mex" staple on every menu? For the uninitiated, they're a rolled tortilla filled with cream cheese and pickled jalapenos then grilled. I don't get it; they're weird

#AI coding can be incredibly powerful, but it's also real bad at coding for maintainability I've been vibe coding a web app, performance was trash—a tab was eating ~2.1GB on load. Turns out it was making ~15 calls for my full DB as part of init, b/c new code is easier. After cleanup, down to ~170MB